🔬 Match packing list PDFs to purchase orders with docling-serve and Ollama

⚡ 45 views · 🔬 Document Extraction & Analysis

Description

Quick overview

This workflow checks an SFTP “pending” folder for packing list PDFs, converts them to OCR text with docling-serve, and uses Ollama (Gemma) to extract structured line items and match them to a purchase order, logging results and moving files to completed or error folders.

How it works

1. Runs every hour and lists files in the SFTP pending folder.
2. Filters for PDFs, downloads each file from SFTP, and logs a PENDING status to an n8n Data Table.
3. Uploads the PDF to docling-serve for OCR conversion (RapidOCR, English) and stops with an error status if conversion fails.
4. Sends the converted Markdown text to an Ollama Gemma chat model to extract a normalized packing-list JSON (PO number, parties, and line items).
5. Loads the matching purchase order data and cross-references PO vs packing-list SKUs and quantities to determine whether the document matches.
6. If SKU count matches but SKU text differs, uses an Ollama-based healing step to detect likely OCR-misread SKUs, applies the corrections, and re-runs the cross-reference.
7. Writes the final status (COMPLETED or ERROR) to the n8n Data Table and moves the PDF in SFTP into a folder named after the status (for example, completed/ or error/).

Setup

1. Create and configure an SFTP credential, and ensure the server has pending/ plus destination folders matching the workflow’s status-based paths (for example, completed/ and error/).
2. Set up docling-serve locally (or update the HTTP endpoint) so http://127.0.0.1:5001/v1/convert/file is reachable from n8n.
3. Configure an Ollama credential and ensure the gemma4:12b-mlx model is available on your Ollama instance.
4. Create an n8n Data Table (or select the existing one) used as document_processor_log to store file name, status, and details.
5. Replace the dummy purchase-order lookup with your real purchase order data source, or ensure the extracted po_number matches one of the provided sample orders.

Requirements

• You will need a (S)FTP server setup with 3 folders: pending, completed and failed
• A n8n data table with 3 columns: fileName, status, details
• Ollama or other AI service
• docling-serve

Customization

• Change the credentials for Ollama in Agent nodes.
• Change the credentials for FTP Nodes
• The move file FTP Node uses the status in lowercases as folder name. If the folder names on the (S)FTP server are not pending, completed and failed then change the set nodes in the Log status and move file section.
• Set your data table in all data table nodes
• Change docling-serve endpoint in HTTP node named “Post PDF to docling-serve”

Additional info

Test files can be found in the github repository of this template

🔗 Nodes Used

HTTP Request, FTP, Schedule Trigger, Filter, AI Agent, Ollama Chat Model

📥 Import

Download workflow.json and import into n8n: Workflow menu → Import from File

📖 Importing guide · 🔑 Credential setup